CHAPEL HILL, North CarolinaMass spectroscopy-based
screening of serum samples from men with elevated PSA levels can distinguish
benign from malignant disease and significantly reduce the need for biopsies,
according to David Ornstein, MD, and his colleagues at the Food and Drug
Administration (FDA) and National Cancer Institute (NCI). Dr. Ornstein is
assistant professor of surgery, Division of Urology, University of North
Carolina School of Medicine. [See Figure]
The work was published in the Proceedings of the 94th Annual
Meeting of the American Association for Cancer Research (abstract 5736). [The
meeting, originally scheduled for Toronto in April, was held instead in
Washington, DC, in July, owing to the outbreak of severe acute respiratory
The underlying data reflect a further development of surface
enhanced laser desorption time-of-flight (SELDI-TOF) mass spectroscopy, a
technology that combines sophisticated computer algorithms with protein binding
chips in order to differentiate proteomic patterns that indicate the presence
or absence of disease.
This approach exploits the fact that diseased organs alter
the proteome (defined as the total array of proteins found in serum) by
addition, subtraction, or modification of the constituent proteins; moreover,
these changes are dynamic, altering as the disease progresses. Only microliters
of sample are required, far less than traditional gel electrophoresis
After application of the sample to the chip, a subset of
proteins will bind, depending on the particular chromatographic matrix
employed; these proteins are then desorbed by laser irradiation, and separated
on the basis of their mass-to-charge ratio, all in a matter of minutes. A
typical profile is shown in the Figure below, accompanied by the view that
would be seen in traditional gel electrophoresis approaches.
By altering the characteristics of the protein chip or the
chromatographic matrix, multiple proteomic profiles can be generated from the
same serum sample. In order for the technique to be a useful diagnostic tool,
it is only necessary to identify specific profiles that reproducibly
differentiate the normal and diseased states.
It is important to note that only a correlation is required,
meaning that there is no need to identify the proteins whose composition is
altered, or to establish any functional link between the changes and the
The approach does require powerful computational resources
and pattern recognition algorithms, however, and imposes strong requirements as
to sample reproducibility, but these constraints have proved tractable. In
practice, the system must first be "trained" by cataloguing samples
from healthy volunteers and from patients who have a given cancer, in order to
identify particular profiles of diagnostic interest.
A schematic representation of the technique is shown on page
1. Typically, 5 to 20 proteins will be involved in discriminating between the
protein signatures of the population of cancer patients and healthy
individuals. Each globular cluster represents the scatter among the patient
population for a given proteomic pattern whose presence or absence has
Individual proteomic signatures from new patients, which
will consist of a single point in the n-dimensional space, are then compared
with these established clusters to make a positive or negative diagnosis. In
the event that the patient’s profile does not fit into any of the existing
clusters, the program will not be able to render an assessment.
Prostate Cancer Study
The technique has already been employed to diagnose ovarian
cancer (Petricoin et al: Lancet 359:572-755, 2002) and prostate cancer (Petricoin
et al: J Natl Cancer Inst 94:1576-1578, 2002). In their most recent
work, Dr. Ornstein and his colleagues concentrated on men with elevated PSA
levels (2.5 to 15 ng/mL), with the aim of reducing the number of follow-up
The system was first trained, using serum from 14 men with 2
or more negative biopsies, and 26 men with biopsy-detected prostate cancer; it
was then applied in a blinded fashion to samples from 136 men, 35 with prostate
cancer, 83 with one negative biopsy, and 18 with two negative biopsies.
Sensitivity of detection was 97.1% (34 of 35 cancers detected) while
specificity was 55% overall.
"These results are very encouraging," Dr. Ornstein
told ONI. "If we had used our algorithm to determine the need for
prostate biopsy among these men with an ‘indeterminant’ PSA level, we could
have prevented more than half of the unnecessary biopsies and would have only
missed one cancer. These results compare favorably to current methods that are
being used to improve the specificity of PSA testing such as percent free PSA
and complex PSA measurement."
An important advantage of the analytic approach is its
adaptive nature, he said; that is, the algorithm is continuously refined as
more data are added.
"Prostate cancer is a heterogeneous disease, and
although prostate cancer is the second leading cause of cancer deaths among US
men, many men are diagnosed with indolent disease and don’t need aggressive
treatment. Our hope is to use this new technology to obtain accurate prognostic
information that can be used to better advise our patients as to the most
appropriate treatment for them," Dr. Ornstein said.
Overall, he said, given further probable technological
improvements, he believes that the proteomic screens will be available at costs
equivalent to other laboratory tests, "and will offer a degree of
precision in diagnosis that is superior to current methodologies," he