| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Guest Access | Sign In via User Name/Password
|
|
|
|||||||
Guest Access | Sign In via User Name/Password |
|||||||||
* From the Dorothy P. & Richard P. Simmons Center for Interstitial Lung Disease (Dr. Kaminski), Pulmonary, Allergy and Critical Care Medicine, University of Pittsburgh Medical Center, Pittsburgh, PA; and the Institute of Respiratory Medicine and Physiology (Dr. Krupsky), Sheba Medical Center, Sheba, Israel.
Correspondence to: Naftali Kaminski, MD, Dorothy P. & Richard P. Simmons Center for Interstitial Lung Disease, Pulmonary, Allergy and Critical Care Medicine, University of Pittsburgh Medical Center, NW 628 MUH, 3459 5th Ave, Pittsburgh, PA 15261; e-mail- kaminskin{at}upmc.edu
Key Words: array comparative genomic hybridization • early detection • microarrays • non-small cell lung cancer • surrogate markers
 |
Introduction |
|---|
 TOP Introduction References |
|---|
Since their introduction approximately 8 years ago, microarrays have transformed biomedical research in general and cancer research specifically. The ability to generate high-resolution genome-scale gene expression profiles allows scientists engaged in basic research to generate new mechanistic hypotheses that should lead to the identification of better therapeutic interventions, while scientists engaged in translational research identify new classes of diseases, correlate prognosis with gene expression patterns, and are close to using the microarray as a diagnostic tool. The most impressive examples of such progress are indeed in cancer research, in which gene expression patterns obtained from microarray experiments allowed the identification of new classes of lymphoma,5 the prediction of metastasis in breast cancer,6 and prognosis determination in lung cancer and in pediatric leukemias and lymphomas.789 In this article, we mainly focus on the translational aspects of applying microarrays to lung cancer research. We present some of the work that has applied microarrays to lung cancer samples and cell lines, and we discuss challenges and future directions.
Tumor Classification: Can We Predict Prognosis Using Gene Expression Patterns?
The TNM staging system for lung cancer is probably the best predictor of survival currently available.10 Despite minor modifications, it has not changed in the last decade and clearly does not account for the prognostic diversity within each stage. The application of microarrays to tumor samples has led to the identification of genes that are associated with different biological behavior of tumors and with prognosis prediction in hematologic malignancies and solid tumors,5681112 suggesting that this is probably the case for lung cancer. Garber et al,13 in a study that was not aimed at looking at prognostic markers, reported the analysis of 67 human lung tumors obtained from 56 patients using complementary DNA arrays. They identified gene expression patterns that were characteristic of every histologic subtype, based on hierarchical clustering algorithms. In adenocarcinomas, they identified three subgroups based on gene expression patterns. These groups seemed to have statistically significantly different survival plots. Unfortunately, 6 of 8 patients in the poor prognosis group (ie, adeno group 3) had evidence of metastases or lymph node involvement (stage IV, 4 patients; stage III, 2 patients),13 while 5 of 11 patients in one good prognosis group (ie, adeno group 1) had stage I disease. This suggests that the statistically significant difference in survival may have resulted from a more advanced tumor stage and not from new molecular classes identified by gene expression patterns. In the same issue of the Proceedings of the National Academy of Science, Bhattacharjee et al14 analyzed 186 lung tumors (127 lung adenocarcinomas, 21 squamous cell lung carcinomas, 20 pulmonary carcinoids and 6 small cell lung carcinomas) and 17 healthy lung samples using oligonucleotide arrays. Similarly to Garber et al,13 they observed that gene expression profiles generally distinguished different tumor types. They identified four adenocarcinoma subclasses, using hierarchical clustering and probabilistic clustering,14 that differed in their gene expression signatures. The C1 cluster contained up-regulated genes associated with cell division and proliferation. In addition to increased levels of genes associated with proliferation, cluster C2 was characterized by increased expression of some neuroendocrine markers. Cluster C3 contained some genes that were in the C2 cluster as well as genes that were seen in the healthy lung and in cluster C4. Interestingly, cluster C4, which was characterized by the increased expression of type II alveolar cell markers, such as TITF1, surfactant proteins SFTPB, SFTPC, and SFTPD genes, and MUC1, contained only stage I tumors. The C2 cluster was associated with a shorter survival time than all other adenocarcinomas (C2 cluster median survival time, 21 months; all non-C2 tumors, 40.5 months; p < 0.005), while the C4 cluster was associated with a better survival time (median survival time, 49.7 months; non-C4 tumor survival time, 33.2 months; p < 0.05). Unfortunately, when these analyses were performed for same-stage patients (stage I), the differences in survival did not reach statistical significance, suggesting that at least some of the distinct gene expression patterns were associated with tumor stage and not with distinct molecular classes in the same tumor stage. Wigle et al15 analyzed 39 tumors using complementary DNA arrays and identified a set of genes that was associated with disease-free survival. However, as in the previous studies, the clinical staging distribution was different between the two groups, with 13 of 24 patients having stage II disease in the early recurrence group compared with 2 patients in the no-recurrence group.
In a study that was better designed to identify gene expression patterns that predict survival, Beer et al7 analyzed 86 primary lung adenocarcinomas (stage I tumors, 67; stage III tumors, 19) and 10 control samples. As expected, gene expression patterns differed significantly between stage I and stage III tumors. In order to identify a robust set of genes that could serve as a real predictor of survival in tumors of similar stages, they applied two equivalent but independent training and testing sets, and "leave-one-out" cross-validation analysis with all tumors. They identified 50 genes that distinguished two groups of stage I lung adenocarcinomas that had a statistically different survival rate. They used this set of 50 genes to identify statistically "good" and "bad" survival groups in all the adenocarcinomas analyzed by Bhattacharjee et al14 and to identify a bad prognosis group in the stage I adenocarcinomas available in this data set. They also verified some of these genes at the protein level using tissue microarrays. While the articles by Garber et al,13 Bhattacharjee et al,14 and Wigle et al15 suggested that there may be survival-related information in gene expression patterns, the work of Beer et al7 clearly demonstrated that a gene expression signature that is predictive of survival in patients with the same stage of disease exists and potentially could be used in the future to guide more aggressive chemotherapeutic treatment of stage I patients who have "bad prognosis" molecular signatures. It is, however, unclear whether such molecular signatures will be more efficacious than a single or few prognostic markers.
Aids to Diagnosis: Can We Use Microarrays To Identify New Tumor Markers?
Although the identification of groups of patients with different prognoses within the same surgical/clinical stage may have significant therapeutic implications, identifying new diagnostic markers for lung cancer could have a significant impact on overall patient survival, on patient management, and on our perception of the disease. Gordon et al,16 in an interesting article that utilized microarray analysis for diagnostic purposes, addressed the difficulty in distinguishing malignant pleural mesothelioma from adenocarcinoma of the lung metastatic to the pleura. They analyzed 181 samples (adenocarcinomas, 150; malignant mesotheliomas, 31) using oligonucleotide arrays. They selected a set of eight genes that had the most significant differences in a training set of 16 adenocarcinomas and 16 malignant mesotheliomas, then they divided the expression levels of the genes that were higher in mesotheliomas by the expression values of all the genes that were lower in adenocarcinomas. When individual ratios were used to classify the rest of the samples (mesotheliomas, 15; adenocarcinomas, 134), they had 91 to 97% accuracy rates. However, when a combined score using three ratios was used, the accuracy rate increased to 99%. They then used real-time quantitative reverse transcription polymerase chain reaction to verify the diagnostic accuracy of the ratios. Expression ratios generated from real-time quantitative reverse transcription polymerase chain reaction correctly diagnosed 96% of the samples.16 This approach may represent a simple and easy way to implement the translation of microarray data into clinical practice. Naturally, testing these diagnostic markers in other malignant mesothelioma data sets will be needed to determine the diagnostic applicability of their data.
An important feature of the use of microarrays and gene expression patterns is protein verification. Of the articles previously mentioned, only one utilized high-throughput protein verification methods to determine the frequency of abnormalities at the protein expression level. One potential approach to the identification of diagnostic markers is a combined use of microarrays and tissue arrays. Sugita et al17 analyzed gene expression patterns in lung tumor cell lines using oligonucleotide arrays and identified members of cancer/testis antigen gene group, including five MAGE-A subfamily members, as diagnostic markers. They then validated the cell line data by immunohistochemical testing of the proteins using a tissue microarray containing 187 non-small cell lung carcinoma samples. The expression of MAGE-A genes correlated with a histologic classification of squamous cell carcinoma, suggesting that they could serve as diagnostic markers.
Another interesting approach to identifying secreted diagnostic markers was recently presented by Welsh et al.18 They analyzed 150 carcinomas from 10 anatomic sites of origin and compared them with 46 healthy tissue samples derived from the corresponding sites of tumor origin, and other body tissues and organs using oligonucleotide arrays. They then used controlled vocabulary terms and sequence-based algorithms that identified genes that encoded secreted proteins. They identified several genes with demonstrated clinical diagnostic or therapeutic application, such as AFP in liver carcinoma, KLK6 and KLK10 in ovarian cancer, and GRP in lung carcinomas. The application of such a systematic analysis to multiple lung cancer microarray data sets should lead to the identification of previously unrecognized serum markers.
Gene Expression Patterns in Peripheral Tissues and Cells as Tools for Early Prediction
So far, all of the work that has dealt with diagnostic and prognostic markers of lung cancer has focused on genes expressed in tumors. However, the main challenge with lung cancer is the early detection of the primary tumor as well as of recurrence. For instance, gene expression patterns in surrogate tissues could help us to determine the identity of current or past smokers who should be screened for the presence of lung nodules by CT scans. Furthermore, the information in gene expression patterns could potentially help us decide which nodules found on CT scan screening need to be resected and which nodules could be safely followed up. While there are very few data to directly support the use of microarrays to identify surrogate markers in peripheral tissues, the rationale is quite clear. Lung cancers arise in the histologically normal lung epithelium as a result of a cumulative process of change in genomic content. The allelotyping of microdissected histologically normal epithelium from smokers with lung cancer revealed thousands of lesions containing clonal abnormalities.19 Furthermore, the gene expression patterns in samples with normal histology from smokers and nonsmokers with cancer seemed to substantially differ in a study limited by sample size,20 suggesting that changes in gene expression patterns associated with cancer in the healthy lungs of smokers could be identified.
Based on these observations, several groups are analyzing gene expression patterns in the bronchial epithelial cells of smokers with and without cancer. We (C. Romano, MSc; I. Shahar, PhD; D. Wilson, MD; and N. Kaminski, MD; unpublished data June 2003) have recently completed a short pilot study of gene expression patterns in peripheral blood monocytes obtained from smokers with and without lung cancer. Preliminary analysis using oligonucleotide arrays of peripheral blood monocytes obtained from five smokers with lung cancer (stage I or II) and five matching control subjects revealed a statistically significant gene expression signature. We are currently actively recruiting more patients in order to verify this observation. It is not completely clear why peripheral blood monocytes should express cancer-specific gene expression signatures. An interesting possibility arises from the observations of the importance active lymphocytic response in early stage non-small cell lung cancer.21 Moran et al22 examined 63 stage I lung adenocarcinomas, and identified the chemokines CCL5, CCL4, and IP-10 to be increased in tumors with significant active lymphocytic response, suggesting that these tumors induce an inflammatory response in circulating lymphocytes and (maybe) that this response could be detected using gene expression profiling methods.
Lung Cancer Biology: Some Future Perspectives and Challenges
The most impressive and obvious observation that one derives when looking at a lung cancer microarray data set is the impressive number of genes that significantly distinguish tumors from tissue samples with normal histology, as well as the number of genes that characterize almost every tumor subclass or phenotype. This observation is not a spurious result of inherent defects in microarray technology. It is a true representation of how different lung cancer cells are from healthy lung tissue. One does not observe these extensive changes when studying nonmalignant lung diseases using microarrays.23 Actually, these extensive and diverse changes in gene expression patterns can almost be predicted from the diversity of chromosomal abnormalities in lung cancer, and from the widespread changes in gene copy number, mutations and altered gene expression regulation by methylation.2 This wide aberration from normal gene expression patterns suggests that, therapeutically, it will be very difficult to design highly effective interventions. However, it does suggest that, diagnostically, early changes in gene expression should be easily identified and quantified.
The availability of such diagnostic gene expression patterns will guide aggressive therapy in patients with early-stage disease, will improve the specificity of current imaging methods for the early detection of lung cancer, and will lead to much better management of lung cancer recurrence. For such methods to be translated into clinical medicine, they need to be robust, and to be based on sound statistics and on the prospective analysis of patient populations. Such methods need to be reproducible between different gene expression platforms as well as to be minimally operator-dependent. Microarray technology needs to mature from the status of a "guerilla" investigational "research-only" technology into a clinical level standardized technology. Currently, this is a significant challenge. On the other hand, the published information about gene expression patterns that are predictive of survival in lung cancer, the identification of new diagnostic markers, and the new and emerging preliminary data on gene expression patterns in surrogate tissues and cells suggest that the quality of information will drive the clinical application. Once markers are prospectively identified, applications will be developed with the potential to transform our perception and management of lung cancer.
|
Footnotes |
|---|
|
References |
|---|
|
TOPIntroductionReferences |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
