| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Commentary |
1 Academic Department of Radiation Oncology, The University of Manchester, Christie Hospital, Manchester M20 4BX, 2 School of Biomedical Sciences, University of Ulster at Coleraine BT52 1SA, Northern Ireland, UK
In the last 10 years, a range of powerful laboratory techniques have become available to measure genes at the DNA, RNA and protein level. This progress in assay methods is associated with a huge increase in the amount of data produced and the development of computational methods for analysing the biological information. These advances have the potential to provide the means of assigning molecular signatures that describe intrinsic differences in how patients respond to radiotherapy. Currently, the methods are being validated as tools that provide prognostic or predictive information on cancer pre-disposition and treatment outcome on an individual basis. In order to raise awareness of recent developments in the area, the British Institute of Radiology's (BIR's) Radiation and Cancer Biology Committee organized a meeting at the BIR on 10 March 2006 to overview the methods and to discuss potential applications for advancing translational research in radiotherapy trials.
The greatly improved opportunities for translational research in radiotherapy trials in the UK were summarized by Professor Peter Hoskin (Mount Vernon Hospital, Northwood). In order to improve the design and quality of trials involving radiotherapy within the UK, trials are now reviewed by the NCRI Radiotherapy Clinical Studies Group. The central review encourages multicentre involvement to facilitate patient accrual. It also allows the identification of potential translational research opportunities at an early stage. There are currently 44 radiotherapy trials open in the UK with more planned, giving widening opportunities for translational questions to be addressed.
High throughput genotyping has the potential to find common genetic variants reported as single nucleotide polymorphisms (SNPs) conferring increased radiosensitivity (Dr Alison Dunning, University of Cambridge). In terms of application in clinical trials, one of the key advantages of genotyping is the relative ease of sample collection. A single blood sample taken in EDTA is required, with some flexibility of the timing of sample collection. An important caveat being that DNA quality and uniformity matters. Although samples are relatively stable compared with those collected for RNA or proteomic analyses, there can be problems with collections involving DNA extracted from mixed types of samples, which can yield unusable data. The HapMap (SNP haplotype map) project has already published its phase 2 results and most of the >2 million documented SNPs in the human genome are now tagged and can therefore be studied. The good news is the technology involved for analysing SNPs has progressed in recent years and high-throughput methods incorporating robotics enable the simultaneous processing of many thousands of samples within a single day. SNP-tagging methods, for reducing the number of SNPs that need to be investigated, are also developing rapidly. Studies can focus on candidate genes likely to be involved in determining an individual's sensitivity to radiation, e.g. genes involved in the recognition and repair of DNA damage. Although each gene of interest might contain several hundred SNPs, the number needed to be studied directly can be reduced by using tagged SNPs. "Tagging" involves identifying regions of the genome (haplotype blocks) where SNPs have evolved together and then finding the SNPs within each block that provide information on most of the other SNPs. Nevertheless, even with candidate genes and SNP tagging, there is an obvious need to collaborate and pool samples obtained in several trials to enable exploration of the full human genome. Since there are many variants, each with probably only a small individual effect, large samples (>2000 patients) are required to have any statistical power to detect real effects due to the variants – the field of association studies has been plagued with false positive results due to small study sizes. For radiotherapy, high-throughput genotyping offers the opportunity to identify a range of radiosensitivity phenotypes, which in the future should allow more accurate tailoring of treatment protocols to a patient's predicted normal tissue radiosensitivity. This could improve treatment planning for all groups of patients from the most radiosensitive to radioresistant.
The potential of cDNA and oligonucleotide microarrays was described by Dr Wendy Allen (Queen's University Belfast). The techniques are now well advanced with evaluation of 60 000 genes per sample possible. The methods are capable of comparing the expressed genes in tumour samples with patient outcome following radiotherapy. A disadvantage of the technology is the requirement for fresh material that is not always readily obtainable within a multicentre trial setting. Methods are now becoming available, however, to extract good quality RNA from wax embedded material which will increase the applicability of the approach, since this type of material is easier to store and much more widely available. Microarrays yield huge amounts of information, and so the bioinformatic handling of the data is crucial for identifying potentially important genes and molecular signatures. Dr Allen emphasised the need for validation studies to verify molecular profiles identified in a training set with a second blinded set of samples. The potential applications of the technique in identifying predictive/prognostic gene signatures was also discussed.
Taking the step from expressed gene to translated protein, Professor Tony Whetton (The University of Manchester) described the current state of the proteomics revolution. Although the technology is continually improving, the capabilities are several years behind the high-throughput DNA and RNA approaches. The techniques required are technically demanding, combining highly sophisticated protein separation methods (two-dimensional liquid chromatography) with mass spectrometry. This is partly improved by new tagging techniques and other approaches which allow relative quantification from several samples as well as protein/peptide identification. One of the biggest problems, however, is the dynamic range issue. The separation of low abundance proteins of interest from high abundance proteins remains critical, although major progress has been made in this area with major protein removal columns.
The preliminary removal of these proteins makes analysis of low abundance, potentially diagnostic proteins much more likely. As gene expression array and proteomic data can be linked, there is also potential to compare transcribed and translated genes in the same samples. Procedures for blood sampling for proteomic studies must be carefully standardized, and the use of proteomic techniques for tissue samples remains problematic and is probably 5 years from being a realistic proposition. Currently, the development of large-scale proteomic studies in radiotherapy trials is not realistic. However, exploration of proteomic techniques in small-scale studies would be of value.
Dr Tim Helliwell from Liverpool University discussed the importance of tissue microarrays (TMAs) and their ability to increase evaluation of multiple samples. With TMAs, several hundred samples can be examined on a single slide. There are several advantages of this approach, although it must be seen as a population screening tool and should not be used for individual patient diagnosis. TMAs allow the relative frequency of a target molecule to be assessed in relation to outcome. It can be used for visualizing proteins (using specific antibodies) or gene amplification/expression (using fluorescent in situ hybridization). The approach is limited by a potential sampling error (which can be minimized by increasing the number of cores per sample), the ability to select the correct probes and also probe availability. However, with more probes becoming available, there is much than can now be done. Again, with the great increase in information there is a need for bioinformatic analysis of the results. TMAs have great potential for obtaining molecular markers associated with prognosis in past, current and future trails.
Dr Francesca Buffa (Gray Cancer Institute, Northwood) highlighted the problem of processing the information that can be produced using the new techniques. The analysis of a large number of genes/proteins and a relatively small number of patients poses several problems regarding variable selection, accuracy of prediction on future data and interpretation of the derived model. Analyses must account for multiple testing and false positives are likely. There are different approaches for handling the data. Supervised methods classify according to a pre-selected clinical end-point – such as locoregional control or disease-specific survival – or biological function. Unsupervised techniques do not require a pre-determined outcome or biological knowledge but cluster genes and/or samples that are similar. A knowledge-based approach can also be used, based on our understanding of biology to look at the expression of genes related to a particular phenotype, such as hypoxia. She discussed the advantages and disadvantages of these approaches from a statistical point of view and raised issues that needed to be considered when designing a trial so that data analysis can be more informative; such as reproducibility and variability of the data collected and standardization of laboratory procedures between centres.
Dr Rob Bristow (Princess Margaret Hospital in Toronto, Canada) emphasised the need for a focused tumour type specific approach to the development of prognostic and predictive biomarker profiles. He described his work on prostate cancer and the evaluation of markers related to radiobiologically relevant phenotypes: hypoxia and activated DNA repair processes. A considerable amount of work is clearly required in evaluating markers in different tumours and those useful in one disease site might have no relevance in another. The potential was highlighted for using phase I or II trial data to develop and explore mechanistic hypothesis applicable to later phase trials, and to establish appropriate times for sampling. There is also a need to evaluate markers not just in clinical trial material but also in relation to the carcinogenic process of different tumours. As some markers might prove to be targets for novel drug development, there is a need to show their expression in tumours compared with surrounding normal tissue.
The importance of being aware of any underlying clinical heterogeneity in biomarker studies was also highlighted (Mr Priy Silva, The University of Manchester). Cancers of the head and neck comprise a heterogeneous group. The multiple sub-sites involved are associated with differences in radioresponse. A study in head and neck squamous cell carcinoma from patients who underwent radical radiotherapy highlighted the potential confounding influence of underlying clinical heterogeneity. Even within a very clinically homogeneous, single sub-site group – oropharyngeal patients who all received radiotherapy to their primary tumour – there was variation in outcome probability in relation to clinical and biomarker data. Differences were shown in the behaviour of tonsil and base of tongue tumours. For example, one marker provided highly significant prognostic information in tonsil tumours and another in base of tongue. Evaluation of potential biomarkers of response must allow for any underlying clinical heterogeneity. Where possible, the powering of studies to enable meaningful subgroup analyses would be useful.
The quality of samples and reproducibility of data obtained in both single and multicentre studies is an important consideration for the design of translational research in clinical trials. The need to establish detailed protocols for sample collection particularly in multicentre studies was emphasised (Dr Catharine West, The University of Manchester). A two-centre study was described involving the collection of head and neck cancer samples during surgery for RNA profiling. All samples taken in RNA later yielded high quality RNA, with significantly greater variation between than within samples. Using a knowledge-based approach for data analysis the derivation of a radiobiologically relevant phenotype was described: a hypoxia metagene, which was shown to yield prognostic information in independent datasets.
The meeting provided no obvious quick-fit answers for what endpoints or approaches to use in the planning of translational aspects of clinical trials. Indeed all of the methods have their advantages and disadvantages and, although they can provide informative data, none can provide all of the answers at present. What did emerge from the meeting was a need for a multidisciplinary and collaborative approach. The development of translational research protocols will benefit from inclusion, at the planning stage, of not just the clinical oncologists involved in running a trial, but also scientists, statisticians, bioinformaticians and pathologists. A strong theme for all the various approaches described was the old adage of "rubbish-in-rubbish-out". Prospective sample collection is clearly important, but must be well thought out with translational research requirements discussed at an early stage in the clinical trial development process. The current funding opportunities for clinical trials and translational research should be exploited by the Clinical Oncology and Radiobiology community and, towards this goal, communication and collaboration are essential.
Received for publication April 7, 2006. Accepted for publication April 13, 2006.
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| BJR | DMFR | IMAGING | ALL BIR JOURNALS |
