Sir

Alizadeh et al.1 show how the analysis of gene expression using high-density DNA microarrays can improve the diagnostic accuracy of diffuse large B-cell lymphoma. But there are many problems to be overcome before this approach may become generally applicable to cancer diagnosis, as hoped for by Berns in his interesting News and Views article “Gene expression in diagnosis”2.

Analysis of gene expression using DNA microarrays is unlikely to replace histopathology as the prime indicator of prognosis. In relation to cancer, the histopathologist makes three important judgements: diagnosing the cell of origin (tumour type) of the cancer; how closely the tumour resembles the tissue of origin (tumour grade); and the extent or site to which the cancer has spread (tumour stage). The extent or stage of the cancer is by far the most important judgement in most types of cancer; it dictates treatment and is an accurate predictor of the prognosis.

Tumour grade also correlates strongly with prognosis, but this is a subjective assessment, which limits its reproducibility and clinical value. It is in grading tumours that microarrays are most likely to improve the accuracy of prognosis.

It is not by chance that the first applications of microarrays to the diagnosis of cancer were made on leukaemias and lymphomas1,2,3. These cancers tend to be single cells that can be obtained non-invasively in a blood sample and can be separated to high purity using cell-surface markers. However, 90 per cent of cancers are solid tumours that have to be surgically removed and are extremely difficult to purify. The microarray approach can provide only a crude average of gene expression across all the cells used to prepare the RNA or DNA. Unless the cancer-cell preparation is highly pure, the contribution of the myriad other cells within the cancer (normal cells, supporting stroma, blood vessels, lymphocytes, and so on) can mask the expression pattern on the array. Pure cancer cells can be obtained by laser microdissection of tissue sections or by cell sorting, but these are labour-intensive and time-consuming processes.

Another problem is the variability in invasive potential between cancer cells within the same tumour. For example, prostate cancer is often present in many distinct foci, only one of which may have the potential to be invasive and dictate the outcome for the patient. Must each focus be analysed separately? Within one focus there may only be a few cells with the potential to invade — will their gene-expression pattern be masked by the surrounding, less malignant cells?

There are already a vast number of prognostic markers available for every type of cancer, and many of these are of independent prognostic significance. However, they are of little use to the individual patient because, although statistically significant, they do not provide the quality of information needed to be confident that a major operation, for example radical prostatectomy or cystectomy, is beneficial.

The acid test for DNA microarrays in cancer diagnosis will be whether the information they provide alters the patient's treatment — a likely outcome but yet to be proven. Using pure populations of cancer cells (for example, cancer-cell lines, flow- or magnetically sorted cancer cells, or cells obtained by microdissection), new and more powerful prognostic markers will be identified using microarrays, as demonstrated by Alizadeh et al.1. It will then be possible to use conventional methods on tissue sections (such as immunohistochemistry and in situ hybridization) to apply these new markers to individual cancers.