with time to spare
William Check, PhD
Discussing his research goal, Richard Roden, PhD, assistant
professor of pathology at Johns Hopkins University, speaks of "our
desperate need" for a good marker to detect ovarian cancer early.
Echoing Dr. Roden, Lance Liotta, MD, PhD, chief of pathology at the
National Cancer Institute and co-director of the FDA/NCI Clinical
Proteomics Program, says, "Ovarian cancer is a disease where there
is great need for an early detection method." Ovarian cancer is usually
diagnosed in its late stages, when disease has spread beyond the ovary.
"If we could switch detection from late to early stage we could make
a tremendous impact on survival," Dr. Liotta says.
David Fishman, MD, expresses a similar view. "We are not doing
our job," says Dr. Fishman, who is professor of obstetrics and gynecology
and director of gynecologic oncology research, Northwestern Memorial
Medical Center, and director of the National Ovarian Cancer Early
Detection Research Network. "We need to be able to find out whether
a woman’s ovaries are abnormal before cancer has spread." The reason
is simple, according to Dr. Fishman—today, as in 1960, less
than 20 percent of women diagnosed with ovarian cancer are alive
five years later. "That to me is unacceptable," he says. "If we
can apply our scientific intelligence clinically to detect ovarian
cancer confined to the ovary, 90 percent of those women will be
alive five years later." In addition, most women diagnosed early
will need much less aggressive surgery and may not require chemotherapy.
In short, Dr. Fishman says, "We need early detection of early
stage ovarian cancer."
If this were the ideal world envisioned by Karl Marx—"To
each according to her needs"—we would have had a marker for
early ovarian cancer many years ago. In our real world, however,
this goal has proved to be elusive.
Steven Skates, PhD, a biostatistician at Massachusetts General
Hospital and assistant professor of medicine at Harvard Medical
School, lists three approaches that have been advocated for early
detection of ovarian cancer in normal-risk women:
In this article we discuss only blood tests, while noting Dr. Fishman’s
caveat: "No one is going to have surgery based only on a blood test. They will
also need an ultrasound imaging test to evaluate the ovaries. A decision on surgery
will come from those combined results."
- An annual blood test followed by an imaging test—typically
transvaginal sonography-for those positive on the blood test.
- Annual imaging, typically with transvaginal ultrasound.
- A blood test and transvaginal ultrasound on all women, now being
investigated as part of NCI’s PLCO (prostate, lung, colorectal,
and ovarian cancer) screening trial.
The best-known serum marker for ovarian cancer,
CA 125, has been studied for many years, but it is not sensitive
or specific enough. A more complex approach, looking at CA 125 levels
over time, appears to be more promising. "Longitudinally following
CA 125 significantly increases preclinical sensitivity," says Dr.
Skates. At the present time, however, CA 125 alone does not suffice.
At best, it detects only 40 to 50 percent of early ovarian cancers.
So researchers are looking at several other markers.
Yan Xu, PhD, associate staff member in the department of cancer
biology at Cleveland Clinic, is investigating lysophosphatidic acid.
Based on preliminary data, Dr. Xu says, "potentially LPA is a big
improvement" over CA 125. Larger trials of LPA are ongoing.
Dr. Roden’s research follows a third approach—attempting
to identify ovarian tumor-associated antigens that are immunogenic.
He has discovered several candidate antigens. "Immunogenic ovarian
tumor-associated antigens might be useful both as biomarkers for
detection and as therapeutic targets," he says.
A fourth distinct tack is being pursued by Dr. Liotta and his
co-director in the FDA/NCI Clinical Proteomics Program, Emanuel
Petricoin, PhD, senior investigator at FDA’s Center for Biologics
Evaluation and Research. Called proteomics, this method seeks characteristic
serum protein expression patterns. Looking at the serum proteome
of ovarian cancer patients was a high priority, says Dr. Petricoin,
"because of its obvious need and its potential ability to make a
"The wealth of information in the serum proteome is astounding,"
says Dr. Liotta. "It contains details and subtleties that tell you
what is going on in internal body sites." He reports that the proteome
patterns that they have observed have been "exquisitely specific,"
differing between ovarian and prostate cancer and between malignant,
benign, and inflammatory conditions. "It is extremely exciting to
think of where we will be in the future with this," he says.
Still, data acquired to date indicate that no one marker—even
the serum proteome—will be adequate. So Drs. Skates and Fishman
and others are collaborating on an NCI-funded study to try to formulate
a panel of markers that is simultaneously sensitive and specific.
Clive Taylor, MD, PhD, chairman of pathology at the University
of Southern California School of Medicine, has written extensively
on tumor markers and what evidence is needed to show they are applicable
to patient care. He agrees that CA 125 has not been specific or
sensitive enough. "It is a bit like the CEA story," he says. "CEA
was initially thought to be a good diagnostic marker for carcinoma
of the colon, but it was not specific or sensitive. Now we know
that it is acceptable for monitoring and followup.
"If you are an optimist," Dr. Taylor continues, "there is some
encouragement that these new looks at protein profiles in serum
are of some real value. But," he cautions, "the same story has been
told before. Fifteen years ago people were making the same claims
about site-directed monoclonal antibodies against peptide products
in serum." Only the appropriate clinical studies will tell whether
the newer tests can stand up to scientific rigor.
Two paramount facts underlie the intense effort to find a biomarker
for early detection of ovarian cancer and the intense difficulty
of this search. As already noted, the motivation is to save lives.
Each year in the United States about 24,000 women are diagnosed
with ovarian cancer, and about 14,000 die of it. This high mortality
rate is due to the fact that at least three-fourths of women are
diagnosed with advanced disease, which has a very low survival rate.
"The fact that most cases are picked up in late-stage disease and
the contrast between early and late-stage survival rates makes early
detection of ovarian cancer an appealing approach to decreasing
mortality from that disease," says Dr. Skates.
However, the annual incidence of ovarian cancer in the target
population—typically considered as all postmenopausal women—is
quite low, about 40-50/100,000. As a consequence, Dr. Skates says,
"Any early detection program has to find one case out of 2,000 or
more women without falsely identifying too many women who don’t
have the disease." Some have suggested that this demands a screening
program with a positive predictive value-the number of women who
have the disease out of those identified as positive—of at
least 10 percent. Many believe it should be higher.
Overall specificity of a screening program must exceed 99.6 percent. "That
is what makes screening for ovarian cancer so difficult," Dr. Skates says. Clinical
trials have shown that it is possible to achieve such high overall specificity
if the specificity of the blood test is around 98 percent, with followup imaging
ruling out most false positives. At the same time, a test should pick out at
least 75 percent of true cases.
While a single assay of CA 125 turned out not to meet these
criteria, Dr. Skates and others have been looking at the value of
measuring CA 125 over time. A statistical model converts the longitudinal
CA 125 profile to a single number called "risk of ovarian cancer,"
which is the risk of having the disease at a given time. One study,
conducted by Ian Jacobs at St. Bart’s Hospital, London, included
22,000 women who had an initial screen with CA 125. Half had three
further annual screens. A retrospective analysis of the data from
this clinical trial indicated that longitudinal CA 125 values increased
sensitivity for ovarian cancer from about 70 percent to 85 percent
while maintaining specificity.
To determine whether this approach increases preclinical detection
of early stage disease—which is the crucial question—Jacobs
is now conducting a randomized trial of longitudinal CA 125 assays
followed by ultrasound compared to annual ultrasound and to a control
Discovery of LPA as a possible biomarker for ovarian cancer grew
out of work in the early 1990s trying to isolate a tumor growth
factor from ascites of patients with ovarian cancer. "Everyone thought
it was a protein," Dr. Xu says, "because known regulators of cancer
cell growth were usually proteins." However, this one turned out
to be a lipid, LPA, which was present in high concentration in ascites.
Further work showed that LPA was also elevated in blood: It was
increased in samples from 47 of 48 ovarian cancer patients, including
nine of 10 with stage 1 disease. "That shows that LPA is a potential
marker for early detection of ovarian cancer," Dr. Xu says. In contrast,
CA 125 was elevated in only two of the nine patients with stage
1 cancer who were tested.
Dr. Xu is continuing to work with LPA, measuring it prospectively
in women coming to Cleveland Clinic for any reason. It continues
to have a high sensitivity and specificity in her hands. However,
she has collected only about 50 samples, since most samples are
being entered into a seven-center clinical trial sponsored by Atairgin
Technologies, the company that has licensed this technique. Atairgin’s
trial is enrolling high-risk women—those who have a suspicious
Dr. Roden’s approach to identifying immunogenic ovarian tumor-associated
antigens uses the recombinant expression cloning method called SEREX,
in which cDNA from a patient’s tumor are recombined into the coliphage
lambda. The sum of the bacterial viruses carrying the human genetic
material is called a cDNA expression library. Each clone’s cDNA
was expressed in bacteria, and the resulting proteins were screened
using antibodies from an ovarian cancer patient.
Five clones whose proteins reacted with the patient’s antibodies
were further investigated. Most efforts so far have focused on two
of these proteins, called HOX A7 and HOX B7. (HOX stands for "homeobox.")
"Homeobox proteins are transcription factors that have very potent
effects during development," Dr. Roden says, "so we sought to assess
the association with ovarian cancer of these two human proteins."
In a preliminary study, Dr. Roden and his colleagues looked at
the prevalence in patients and controls of antibodies reactive with
HOX A7 or HOX B7. For HOX A7, sera from healthy women were minimally
reactive, while sera from 16 of 24 women with moderately differentiated
ovarian serous carcinoma had reactivity. Reactivity to HOX A7 was
also present in serum from 13 of 19 patients with benign serous
cystadenoma and one of 24 with poorly differentiated ovarian carcinoma.
For HOX B7, serum from 13 of 39 ovarian cancer patients and one
of 29 healthy women were reactive. One complication was that HOX
B7 RNA expression was elevated in all 15 ovarian carcinomas
tested, even though not all patients made anti-B7 antibody. "It
looks as though in some cases a patient can have elevated expression
without antibody formation," Dr. Roden says.
He is also intrigued by finding strong antibody levels to HOX
A7 even with localized disease, such as benign serous cystadenoma.
On the negative side, that could reduce the test’s ability to distinguish
benign from malignant ovarian disease. On the positive side, Dr.
Roden notes, "That observation holds promise for detection of localized
carcinoma." However, he says, "We have not yet tested blood samples
from patients with early stage ovarian carcinoma—which is
really what is wanted."
Detecting potential biomarkers by measuring the immune response
to tumor-associated antigens rather than tumor-associated antigens
themselves has several advantages, Dr. Roden says. One is that the
test for antibodies in a patient’s serum is very simple. Another
is that the tumor-associated antigen does not have to be released
from the tumor to stimulate an immune response. "Most important,"
Dr. Roden says, "recognition of an immune response represents a
powerful amplification of the signal—the tumor-associated
antigen—that the tumor presents."
He is continuing to screen for new antigens with an improved methodology.
Expression of antigens in bacteria through the lambda system does
not allow for posttranscriptional modification of proteins, such
as glycosylation. "Such changes may occur in tumors and be recognized
by antibodies," Dr. Roden notes. He is now using patient antibodies
to precipitate tumor antigens, which are then identified by mass
Amplistar Inc. is validating the utility of the five immunogenic
ovarian cancer gene products for diagnostic purposes by conducting
a larger case-control series.
Dr. Roden says that it is still "early days" in the assessment of the clinical
value of this approach.
To identify proteomic patterns characteristic of ovarian
cancer, a computer algorithm is employed. Dr. Liotta explains that
the first step is to scan the thousands of proteins in a drop of
serum with mass spectroscopy looking at varied subsets of proteins
to detect any reproducible pattern of difference between normal
and ovarian cancer. However, the computer must first be trained
to recognize such patterns by showing it a training set of samples.
"In training the computer, selecting the training set is probably
the most important step," says Dr. Petricoin. "These types of AI-based
machine learning systems are not like having HAL on the other side
of the computer screen saying, ’Hello, Dave,’" he notes. "These
are dumb systems. If you don’t show it the right data set, it can’t
For ovarian cancer, the training set comprised 50 blood samples
from women with various stages of ovarian cancer as well as 50 samples
from healthy women—those who had been followed for five years
after the serum sample was taken and had no evidence of ovarian
cancer, although some did have ovarian cysts.
Once the algorithm could distinguish benign from malignant samples
with complete accuracy, it was shown a test set of 50 samples from
women with ovarian cancer and 50 from women free of malignant disease,
all distinct from the training set. In addition, the testing set
included 16 samples from women with a variety of nonmalignant disease
conditions, such as sinusitis and rheumatoid arthritis. "It is important
to show that your system can discriminate malignant from benign
disease," Dr. Petricoin says. In a general population, nonmalignant
conditions such as ovarian cysts, fibroids, endometriosis, and general
inflammatory conditions will be seen much more frequently than ovarian
cancer. Attaining adequate specificity demands that the algorithm
can distinguish between ovarian cancer and these other disorders.
In the test set, the algorithm correctly identified all cancers,
including 18 stage 1 cases. It erroneously classified three of 66
(4.5 percent) benign samples as cancerous. These results yield a
sensitivity (100 percent) exceeding the criteria cited by Dr. Skates,
but a specificity (95 percent) that falls short of the 98 percent
benchmark. A 4.5 percent false-positive rate is statistically impressive.
However, given the low prevalence of ovarian cancer, among 20,000
screened women a 4.5 percent false-positive rate translates into
1,000 women wrongly sent for further workup to detect 10 who have
ovarian cancer. "This is not exactly what we wanted," Dr. Petricoin
concludes. "There is still some work that needs to be done." He
notes that these numbers would be more acceptable in a high-risk
population, so initial clinical testing will be done in women at
increased risk of ovarian cancer.
In a refinement of the algorithmic approach, these investigators
have subsequently evaluated combinations of models. Dr. Petricoin
explains that the model that they published was only one of 12 potential
models derived from the initial analysis of the spectroscopic datastream.
They attempted to combine two or more models to increase specificity
while maintaining sensitivity. Selecting the optimal combination
of models is crucial. "If each model gets a different subset wrong,
all you do ultimately is to get more wrong with the combination,"
Dr. Petricoin says. Eventually they did find a combination of models
that was completely sensitive and specific for the test data set.
Validating a combination of models is the next step. "We are still
trying to lock on to the best combination of models," Dr. Petricoin
says. "We are excited to actually test this model in a true clinical
setting." Later this year they anticipate opening the world’s first
clinical reference laboratory for serum pattern proteomic diagnosis,
which will handle serum samples for ovarian, breast, and prostate
cancer in a clinical trials context. Analyses will be done in Dr.
Liotta’s laboratory, which has CLIA and CAP licenses and provides
clinical test results. "We will run all of these samples under a
rigorous QC/QA process that we are developing to generate very reproducible
results," Dr. Petricoin says.
Initially the combination of models will be evaluated in a clinical
trial of high-risk women that will be supported by NCI and FDA.
Subjects will be women with one of three risk factors for ovarian
or breast cancer. One group will be women who are being followed
for recurrence. A second group will be women who are at increased
risk of ovarian and breast cancer because they have a first-degree
relative with breast or ovarian cancer or because they have mutated
BRCA genes. These women are being followed in high-risk clinics.
"This is a very anxious and worried group of women," Dr. Petricoin
says. "Some are choosing to have prophylactic oophorectomies without
any medical justification to get rid of their anxiety because they
saw their mother die of ovarian cancer."
In the third group will be women with a pelvic mass. In these
women, serum will be collected before surgical analysis to see whether
the proteomic pattern can distinguish benign from malignant cases.
Pelvic masses seen on ultrasound turn out to be ovarian cysts or
fibroids a very large percent of the time. Since prudent management
dictates waiting two to three months to see whether the mass spontaneously
resolves, there is a delay in the patient seeing a gynecologic oncologist,
after which a laparoscopy may be done. "When we are talking about
ovarian cancer, a delay of two to three months could translate into
a real difference in survival probability," Dr. Petricoin says.
"In that population, we might risk more false positives to get those
with true cancer to be seen by a gynecologist oncologist as soon
In the clinical trials, proteomic testing will be done without
charge to patients. Once the method is approved, any company can
commercialize it using data gathered in the clinical trial for PMA
filing and obtaining a commercial license from Correlogic Systems
Inc. Dr. Petricoin says the proteomic method presents "a unique
situation." License to the underlying intellectual property is held
jointly by the U.S. government and Correlogic Systems, a software
company that developed the basic algorithms and holds exclusive
commercialization rights. Correlogic Systems and the U.S. government
are coinventors and licensees of the patent.
Both Dr. Liotta and Dr. Petricoin express caution about how the
proteomic technique is applied. "Right now we can’t envision this
as being a general population screen for ovarian cancer," Dr. Liotta
says. "Even if we had 99 percent sensitivity and specificity, we
would have wrong results in one percent of persons screened. We
couldn’t tolerate that level of misdiagnosis in a screen of millions
of women. We don’t want to make any mistakes."
"It would be very dangerous to think about this as a general screening test,"
Dr. Petricoin concurs. "That is where many biomarkers have failed—zealous
investigators pushed their favorite biomarkers as clinical screening tests.
Just look at the conundrums that mammography is creating right now. We don’t
want to increase confusion, but to try to help physicians make diagnoses."
Dr. Skates suggests that maximizing early stage preclinical
sensitivity at a fixed specificity of 98 percent will require looking
at multiple markers. First it will be necessary to work out a way
to translate time series of measurements for one or more markers
into a risk value. Then the level of risk can be set for a specificity
of 98 percent. "We need to try both to identify markers that will
maximize sensitivity and to find a way to combine that information
into one number," Dr. Skates says.
Right now Dr. Skates and colleagues are comparing preoperative
serum in cases to serum from control subjects. "We are not ready
to implement this approach in a prospective trial," he says. "But
through statistical analysis of data we are trying to identify the
best panel of biomarkers to give optimal sensitivity for the target
specificity." One question they are addressing is how many markers
to use. "I don’t think there is an agreed-upon number of markers,"
Dr. Skates says. "It will be a trade-off between complexity of method
and interpretation and diminishing returns from adding a new marker
to the panel. Our hope is that these complex methods will be built
into a system that measures multiple markers over time and summarizes
the information in a single number that indicates the patient’s
risk of having the disease."
Dr. Fishman, who is one of the principal investigators on this
NCI-funded study, says that it will include both general and high-risk
populations, as well as subjects with cancer. "We will evaluate
biomarkers independent of each other and in combinations," he says.
"We already have a fair amount of preliminary data showing that
these markers are real. Now we are trying to validate them in larger
populations—thousands of women—and to compare and combine
Dr. Skates emphasizes that, for now, "All of these results must
be viewed in the context of a research study only, not for general
clinical use, since none of these tests has been shown to have an
impact on mortality." He adds, "We want to temper our enthusiasm
to looking at these approaches through research studies—to
determine their effect before advocating them for the general population."
As to when definitive data might be available, Dr. Skates says,
"I am always very tentative when it comes to predictions, but I
would hope we can identify a promising candidate panel by the end
of the year and incorporate it into a prospective trial. Or at least
begin the process of incorporating it into a prospective trial,
which may be a pilot trial, sometime next year."
Dr. Fishman has a more ambitous goal for the next year. "My hope,"
he says, "is that a year from now we will have the answers. I anticipate
at least two of the blood tests—which are in FDA evaluations
right now—will be approved by this time next year. It would
be absolutely fantastic if that would be the case."
William Check is a medical writer in Wilmette, Ill.