Genomics and proteomics: expression arrays in clinical oncology

Genomics and proteomics: expression arrays in clinical oncology

Annals of Oncology 15 (Supplement 4): iv163 – iv165, 2004 doi:10.1093/annonc/mdh921 Genomics and proteomics: expression arrays in clinical oncology M...

38KB Sizes 0 Downloads 25 Views

Annals of Oncology 15 (Supplement 4): iv163 – iv165, 2004 doi:10.1093/annonc/mdh921

Genomics and proteomics: expression arrays in clinical oncology M. F. Fey Institute of Medical Oncology, Inselspital, Bern, Switzerland

Introduction

q 2004 European Society for Medical Oncology

Molecular diagnosis of tumour entities with the chip technology The distinction between acute myeloid (AML) and acute lymphoblastic leukaemias (ALL) is traditionally made by morphology, immunological markers and selected molecular analyses. The diagnostic work-up is therefore fairly laborious and involves a number of different techniques and strategies. One wonders whether chip analysis of gene-expression profiles would provide an all-in-one approach to this problem. It turns out that gene-expression profiles can be defined which neatly distinguish AML from ALL [2], or which help to identify mixed-lineage leukaemias (MLL) with alterations of chromosome 11q23 as a distinct clinico-pathological entity [4]. However, the chip technology still requires a lot of technical and logistical polishing before we may even think of putting the light microscope and the flow cytometer aside in any haematology laboratory. In non-small-cell lung cancer samples, panels of genes have been found which differ in their expression from normal lung tissue, and thus provide a set of markers that may be useful as diagnostic or therapeutic targets. Profiles seem to follow histopathological classification very closely, and therefore chip analysis is unlikely in the near future to replace the microscope in lung cancer diagnosis. However, tumour-specific expression levels of distinct groups of genes may provide

Downloaded from http://annonc.oxfordjournals.org/ at Florida International University on July 1, 2015

Molecular profiling of human cancer is the large-scale analysis of gene expression using the new DNA array technology. DNA microchips consist of rows and rows of oligonucleotide or cDNA strands lined up in dots on a miniature silicon chip or glass slide surface. Oligonucleotides correspond to short segments of DNA complementary to mRNA transcripts. cDNA arrays are constructed from mRNA libraries prepared from cells of interest, representing their patterns of gene expression. In principle, DNA microarrays are also suitable for detecting genomic alterations in DNA samples, but current interest in cancer focuses largely on transcriptional profiling of cancer cells. In 1994, the first commercially available microarray, produced by Affymetrix, accommodated 16 000 probes representing a few thousand genes. The most recent arrays now contain up to 1.5 million features. Arrays can therefore accommodate up to around 30 000 oligonucleotides or cDNA sequences, and the most modern ones detect the expression of nearly 50 000 transcripts using over 1.3 million probes. The selection of these sequences or probes may either be random, or specialised, i.e. relevant to a particular tissue (‘lymphochip’) [1] or a biological process. To use arrays for the detection of transcriptional profiles in cancer samples, RNA is extracted from tumour tissue, labelled and hybridised to the array. Oligonucleotide microarray hybridisation of a sample provides a direct estimate of the abundance of a given RNA transcript. In contrast, spotted cDNA microarrays usually require sample hybridisation in the presence of competitor reference RNA. Accordingly, information on gene expression is obtained as a relative concentration (ratio) of a transcript with respect to the reference sample (mostly tumour versus selected reference tissue). Gene-expression levels provided by chips can then be compared across many samples, normal and pathological. The choice of reference RNA is crucial to these analyses. For example, in myeloid leukaemia, one might consider a comparison with lymphoid leukaemias, with ‘normal’ granulocytes, normal bone marrow, or with selected normal CD34+ haematopoietic stem cells [2, 3]. Accordingly, gene-expression profiles detected in leukaemic cells will differ with respect to various reference samples. The task of gene profiling is by no means completed at the point where raw data from chip hybridisation become available, since these data need appropriate sorting. With suitable pattern recognition software, groups of genes can be identified which are

expressed in parallel, and thus (tumour) samples can be clustered together by virtue of expressing a similar repertoire of genes, i.e. a specific profile or gene expression signature. Clinical data can then be superimposed, for example through separation of tumours into two or several clinically relevant categories (e.g. high grade non-Hodgkin’s lymphoma either cured or fatal; so-called supervised clustering) [3]. The statistical analysis of huge chip data sets still remains a major challenge, usually beyond the immediate grasp of many clinicians or biologists interested in cancer. Specialised software for data analysis is increasingly becoming available, but it is advisable to seek the advice of well-trained bio-statisticians before any chip experiments are planned or executed. Failure to do so at an early stage may well result in a sobering discussion with a statistician or a data analyst of how best to prepare a funeral of the project. As clinical oncologists we are most interested in the clinical application of this new technology. A vast amount of literature in this field is accumulating relatively rapidly, to be illustrated by a few selected examples.

iv164

Predictive and prognostic tumour markers detected through chip analysis Although a relatively large number of prognostic and predictive markers are now available in breast cancer, the selection of women for different adjuvant treatments based on their risk profile is still somewhat unsatisfactory. This is illustrated by a quick glance at the latest edition of the St Gallen Consensus Conference [9], which (i) splits breast cancer into relatively crude categories (node-negative versus node-positive) and (ii) offers general guidelines for treatment that are still vague and provide much room for arbitrary clinical decisions, for example on the type of adjuvant chemotherapy to be given. Gene-expression profiles of tumour material from women with early stage node-negative disease seem to delineate clinically relevant subgroups with markedly diverse gene-expression patterns, which correlate with outcome [10, 11]. It is still not possible to draw any firm conclusions as to how one should tailor treatment to these ‘new’ risk categories, but a basis for progress in this direction has now been established. About 50% of patients with diffuse large B-cell lymphoma are cured by currently available chemotherapies. Despite the use of the International Prognostic Index (IPI), clinical data, morphology and other tools still offer too few prognostic factors to separate all cases that will do well from those with an eventually fatal course. An algorithm based on expression profiles was able to characterise two categories of high-grade lymphoma patients with very different 5-year survival rates (70% versus 12%), most of them undistinguishable by morphology and immunological markers [3]. Whilst at present these data do not allow us to tailor treatment to these new lymphoma entities, as such, they do provide a basis for a future more rational targeted and individualised approach to the therapy of high-grade lymphoma.

Prediction of treatment results in cancer drug therapy Combination chemotherapies in cancer are usually selected by a trial-and-error approach. It should now be possible to use

gene-expression profiling to assess the interaction of anticancer agents in order to optimise combination chemotherapy. In ALL, a comparison of samples before and after treatment with typical anti-leukaemic agents (methotrexate and mercaptopurine) yields a set of genes with distinct regulation patterns linked to treatment [12]. Gene-expression responses after therapy may therefore be drug-specific, and the results of combination chemotherapy may thus add up to more than the sum of single agent effects. This concept has long been known based on clinical observations, but the chip technology has now provided the tool to look more closely into its molecular basis. Chip analysis has shed light on anti-androgen resistance in the treatment of prostate cancer [13]. It appears that increased androgen-receptor expression is the defining factor in the process of prostate cancer cells turning hormone-refractory. Modest changes in the level of androgen-receptor protein might upset the balance of co-factors that regulate transcription of androgen-receptor target genes. Hence, when more androgenreceptor protein is present, antagonists can no longer inhibit transcription, leading to clinical therapy resistance. A comparison of rectal cancer biopsies taken before and during treatment with 5-fluorouracil-based regimens neatly demonstrates that a set of genes required for RNA and protein synthesis, and for metabolism is down-regulated in cancer cells, but not in normal colorectal mucosa [14]. Thus, gene-expression profiling has the potential to enhance the development of anti-cancer agents, and also to promote our understanding of their mode of action in the clinical setting. In non-small-cell lung cancer, expression profiles can be defined that predict the chances of successful treatment with commonly used anti-cancer drugs such as cisplatin, taxanes or gemcitabine [7]. If confirmed on a larger clinical scale such profiles would eventually permit selection of patients with a particularly high likelihood of responding, and spare those patients who most likely will suffer the side-effects of chemotherapy, but will not benefit.

Conclusion In summary, DNA microarrays provide a significant boost to cancer research, both in basic experimental and translational projects [15 –18]. With more refinements of the technology on the horizon, its broader clinical application might become a reality. It is, however, too early to incorporate the chips into daily diagnostic routine, since a fair number of important issues regarding their proper clinical use are still unresolved. These are technical (quality control and validation), financial (expenses prohibitive at this time, although costs are coming down), logistical (need of ready-to-use and simple software for data analysis), and above all clinical. Even if we may nowadays split heterogeneous cancers, such as breast cancer or lymphoma, into many molecularly defined entities, we do not have established targeted treatments for all of these patient subgroups. The path from molecular genetics to individualised treatment is still a long one, characterised by incremental

Downloaded from http://annonc.oxfordjournals.org/ at Florida International University on July 1, 2015

useful markers for the development of targeted treatment regimens [5 –7]. Carcinoma of unknown primary site is a well known problem in clinical oncology. In spite of extensive work-up by experienced pathologists, the origin of many cases remains unclear, and the choice of drugs to treat such patients must therefore be relatively broad to cover a variety of possible tumour types. Transcriptional profiling may help [8]. In a study of ovarian, breast, colon and pancreatic cancers, sets of genes were identified, which allowed allocation of >80% of samples to their correct diagnostic domain, and profiles seen in metastases clustered tightly with those seen in the corresponding primary tumours. Such an expression map may provide a reliable and practical approach to determine the tumour type, and hence the primary site, in patients with metastatic carcinomas of clinically unknown origin.

iv165

References 1. Alizadeh AA, Eisen MB, Davis RE et al. Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature 2000; 403: 503 –511. 2. Golub TR, Slonim DK, Tamayo P et al. Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science 1999; 286: 531–537. 3. Shipp MA, Ross KN, Tamayo P et al. Diffuse large B-cell lymphoma outcome prediction by gene-expression profiling and supervised machine learning. Nat Med 2002; 8: 68–74. 4. Armstrong SA, Staunton JE, Silverman LB et al. MLL translocations specify a distinct gene expression profile that distinguishes a unique leukaemia. Nat Genet 2002; 30: 41–47. 5. Virtanen C, Ishikawa Y, Honjoh D et al. Integrated classification of lung tumours and cell lines by expression profiling. Proc Natl Acad Sci USA 2002; 99: 12357–12362.

6. Heighway J, Knapp T, Boyce L et al. Expression profiling of primary non-small cell lung cancer for target identification. Oncogene 2002; 21: 7749–7763. 7. Kikuchi T, Daigo Y, Katagiri T et al. Expression profiles of nonsmall cell lung cancers on cDNA microarrays: identification of genes for prediction of lymph-node metastasis and sensitivity to anti-cancer drugs. Oncogene 2003; 22: 2192– 2205. 8. Buckhaults P, Zhang Z, Chen YC et al. Identifying tumour origin using a gene expression-based classification map. Cancer Res 2003; 63: 4144–4149. 9. Senn HJ, Thu¨rlimann B, Goldhirsch A et al. Comments on the St. Gallen Consensus on the primary therapy of early breast cancer. Breast 2003; 12: 569 –582. 10. Van t’Veer LJ, Dai H, van de Vijver MJ et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002; 415: 530–536. 11. Perou CM, Serlie T, Eisen MB et al. Molecular portraits of human breast tumours. Nature 2000; 406: 747–752. 12. Cheok MH, Yang W, Pui CH et al. Treatment-specific changes in gene-expression discriminate in vivo drug response in human leukaemia cells. Nat Genet 2003; 34: 85 –90. 13. Chen CD, Welsbie DS, Tran C et al. Molecular determinants of resistance to anti-androgen therapy. Nat Med 2004; 10: 33–39. 14. Clarke PA, George ML, Easdale S et al. Molecular pharmacology of cancer therapy in human colorectal cancer by gene expression profiling. Cancer Res 2003; 63: 6855–6863. 15. Ramaswamy S, Tamayo P, Rifkin R et al. Multiclass cancer diagnosis using tumour gene expression signatures. Proc Natl Acad Sci USA 2001; 98: 15149–15154. 16. Fey MF. Impact of the Human Genome Project on the clinical management of sporadic cancers. Lancet Oncol 2002; 3: 349–356. 17. Aitman TJ. DNA microarrays in medical practice. BMJ 2001; 323: 611–615. 18. Mohr S, Leikauf G, Keith G, Rihn BH. Microarrays as cancer keys: an array of possibilities. J Clin Oncol 2002; 20: 3165–3175.

Downloaded from http://annonc.oxfordjournals.org/ at Florida International University on July 1, 2015

steps rather than big leaps. A list of genes resulting from a microarray survey should not be regarded as an end in itself. Its real value appears only as genes on that list move through the process of biological validation. Tools that can automatically indicate the functional importance of particular geneexpression profiles, and link them to biological pathways, are still in their infancy, and much work awaits experimental and clinical researchers to exploit the wealth of information provided by chips in the best possible way. Whilst these caveats are still appropriate at present, it would be unfair to conclude this survey with a table of criticisms and a to-do list indicating the remaining problems of the chip technology. The insights into cancer biology and the clinical concepts that have emerged from studying cancer with DNA microarrays are considerable, and likely to change, or at least colour, our way of understanding cancer for a long time to come.