The author discusses the need for an
accurate yet simple test for ovarian cancer and how investigators are
exploring novel, rapid ways of detecting the disease. In their search,
new, sensitive technologies for molecular analysis – primarily mass
spectrometry-based techniques – have been developed
| Mar
31, 2005 |
| By:
Conan
Li |
| Pharmaceutical
Discovery |
|
Ovarian cancer is an insidious disease that kills more than 15,000
Americans each year (1). The lethality of this disease stems from our
inability to diagnose it easily; this is because its symptoms — such
as nausea, loss of appetite and abdominal discomfort — are common to
many disorders. Most women are diagnosed with ovarian cancer in
late-stage disease and have a five-year survival rate of less than 30%
(1). Yet, survival rates soar to over 90% if the disease is discovered
when cancer still is localized to the ovaries (1).
Table I. Women in the U.S. at high
risk for ovarian cancer
|
Unfortunately, current methods of diagnosis, such as transvaginal
ultrasound, laparoscopy or positive emission tomography are impractical
for general testing; they are complex procedures that only can be
performed with a visit to a medical facility. Even applying these
procedures to just the nearly 10 million women in the United States who
are at high risk (Table I) would pose a tremendous burden on the
healthcare system. The cancer biomarker CA-125 is not appropriate for
screening as it detects ovarian cancer only half of the time (2). The
need for an accurate yet simple test has prompted investigators to
explore novel, rapid ways of detecting ovarian cancer. In their search,
new, sensitive technologies for molecular analysis have come to the
fore.
Figure 1. Each serum sample in the
Zhang study was processed into six fractions and run in
triplicate on four types of chips, for a total of 72 spots per
sample (13).
|
To be practical for testing the high-risk population, a rapid detection
assay for ovarian cancer must meet certain criteria. The assay must be
able to be performed on an easily obtained specimen, such as a blood
sample, so women can be tested routinely and conveniently (Table II). An
assay that requires a tissue biopsy, for example, would be ruled out as
a convenient test. The assay must be robust, such that normal handling
and transport of the specimen to the testing laboratory does not alter
the analyte or biomarker being measured. For example, in routine
laboratory testing, a sample is drawn at a patient service center and
can be stored at room temperature or normal refrigeration for several
hours before a courier packs it in dry ice or a cold pack for shipment.
The sample could take a day or longer to arrive at the laboratory. The
need for special sample processing not normally performed at the patient
service center, such as flash-freezing or extraction of a particular
molecular component, could make it difficult to adapt the test for
routine use. In the lab, the process for testing the sample should be
high-throughput and automated.
Figure 2. A schematic of the
experimental protocol for lysophospholipid (LPL) analysis to
determine which lipid biomarkers correlate with ovarian cancer
(16).
|
Furthermore, if only a handful of
laboratories could run the test, it is estimated that testing the
high-risk women alone would require more than 5000 specimens to be
processed per day at each location. Automation is key: not only is
skilled laboratory labor expensive, but also manual specimen handling
can cause sample mix-ups and other errors. The accuracy requirement of a
detection test is a matter of some debate. Because ovarian cancer is
rare — there are 26,000 new cases in the U.S. per year (1) — even a
sensitivity of 99% would imply that 100,000 of the high-risk patients
would test positive but not have cancer. The positive predictive value
of such a test would be less than 20%. Does this mean the accuracy
requirements for an ovarian cancer detection test would be so high that
no assay could ever meet it? It's important to consider that there is
not even a simple assay that can point out which women should undergo
more definitive testing; such a test would provide value. In other
words, for ovarian cancer there is no counterpart to the Pap smear, used
to help screen for cervical cancer.
The multiple ovarian cancer
detection tests under development are based upon different,
complementary technologies and disparate biomarkers, so in principle
their combined use will provide higher accuracy. Suboptimal sensitivity
of a detection assay can be compensated somewhat through regular testing
of women at high risk; a convenient assay makes such routine testing
less burdensome and increases patient compliance. These arguments
suggest that an assay with even 95% sensitivity and specificity should
help in the management of ovarian cancer.
Table II. Rapid detection of
ovarian cancer: Laboratory requirements
|
The importance of developing a rapid
detection assay for ovarian cancer has spurred technological advances on
many fronts. Scientists assume the definitive biomarkers would be of low
concentration, so sensitive techniques such as mass spectrometry (MS)
are logical tools. The first indication that rapid detection of ovarian
cancer might be within our grasp was the study by Petricoin et al., (5)
using surface-enhanced laser desorption ionization time-of-flight mass
spectrometry (SELDI-TOF MS) (Ciphergen Biosystems, Fremont, California,
USA). In SELDI-TOF MS, a protein sample is spotted onto a chip that has
been functionalized with chemical groups that attract molecules based
upon their charge properties. An ionization enhancing material,
a-cyano-4-hydroxy-cinnamic acid, is deposited on the protein spot. The
ionized molecules, desorbed from the sample with a laser beam,
accelerate in an electric field and migrate through a flight tube to the
detector. The time it takes a molecule to traverse the flight tube is
proportional to the square root of the mass-to-charge ratio, M/Z
(6). Thus, molecules desorbed from the sample substrate at the same time
will arrive at the detector at different times and diverge on the basis
of size and charge. TOF results are displayed as a mass spectrum,
showing molecular intensity as a function of M/Z. Using SELDI-TOF,
Petricoin et al., analyzed serum samples from cancer and non-cancer
subjects. The spectra were analyzed using a genetic algorithm developed
by Correlogic Systems Inc. (Bethesda, Maryland, USA), in which random
sets of M/Z values, or markers, are used to distinguish the
cancer and non-cancer spectra. Sets of markers that successfully
distinguish the spectra are considered further while poor patterns are
discarded, until the optimal patterns are found. This algorithm,
therefore, is termed, "genetic." The spectra of 216 samples
(100 cancer and 116 non-cancer) were analyzed to find the best pattern
that was capable of distinguishing cancer from non-cancer samples with
100% sensitivity and 95% specificity. In December 2003, Zhu et al., at
SUNY (State University of New York, Stony Brook, NY, USA) (7) used a
different approach to analyze the ovarian cancer spectra posted by the
National Cancer Institute (NCI, Bethesda, Maryland, USA) on the Internet
(8). They smoothed and normalized each mass spectrum, then compared the
cancer and non-cancer spectra to identify molecular markers (M/Z
positions) with intensities that differed significantly between the two
sets. Using a randomly selected training set of 50 cancer and 50
non-cancer spectra, their analysis yielded a pattern of 18 markers that
could classify or distinguish an independent set of 369 spectra into
cancer and non-cancer with 100% accuracy. Zhu et al., repeated their
process another 50 times with randomly chosen training spectra and each
time obtained patterns that could classify an independent set of spectra
with 100% accuracy. There were no overlapping markers in the published
discriminating patterns from the Petricoin and Zhu studies (5, 7),
indicating that the approaches need further refining and that other
evidence may have to be considered for optimal biomarker panel
selection.
Figure 3. Combined 16:0 LPA and
20:4 LPA levels in plasma from pre-operative ovarian cancer and
control subjects (16). Reprinted with permission from the
American Association of Cancer Research.
|
A very fundamental difference exists
between the spectral analysis methods used by Correlogic Systems and
SUNY. While Correlogic Systems' genetic algorithm treats each of the
15,200 M/Z positions in the SELDI-TOF spectrum as a potential
candidate for the discriminating marker set, the SUNY approach considers
only peaks with magnitudes substantial enough to have survived a
Gaussian smoothing routine and still distinguish cancer and non-cancer
spectra. Thus, SUNY's approach singles out abundant molecules and is
more likely to yield markers with biological significance. Additionally,
in the Correlogic approach results could be more susceptible to slight
mass shifts from run to run, because spectral registration was not
addressed, except by once-daily calibration (5). The spectral smoothing
step employed by the SUNY group in essence accommodates mass shifts of
up to 0.2%, thereby enabling spectrum-to-spectrum registration and
comparison (7). Mass spectrometry pattern analysis is a potentially
rewarding approach in that it utilizes the power of combined multiple
markers so that, in principle, discrimination accuracy is higher. Having
reliable, discriminatory patterns obviates the need to identify and
purify the markers of interest and develop molecular assays for them.
This process can be quite tedious, especially if the markers are in low
concentration. Furthermore, mass spectrum pattern assays take advantage
of the high resolving power and small sample volume requirement of mass
spectrometry.
One drawback to the mass spectrum
pattern method is its lack of precedence in the laboratory, prompting
the need for more rigorous and creative means of eliminating artifacts
and validating performance (9). Petricoin's study has not been without
its detractors. For example, Sorace and Zhan (10), using Wilcoxon
Research Inc. (Gaithersburg, Maryland, USA) statistics, analyzed the NCI
data gathered on August 8, 2002 and found the majority of markers in the
panel they derived to have an M/Z of less than 500. From this
they contended that the cancer and non-cancer spectra were
distinguishable by a "non-biologic experimental bias (10)."
However, they did not analyze the data in the Petricoin publication, and
Zhu's analysis of the same August 8 data set yielded markers with higher
M/Z values and potential biological significance.
Mass spectrometry pattern analysis
also raises novel, complex issues in terms of regulatory oversight.
Though assays that have been developed and validated in reference
laboratories normally can be used for clinical testing without FDA
clearance, the software used in pattern development and analysis could
fall under FDA regulation, as in the case of the Correlogic software
(11). The mass spectrum pattern method also requires laboratorians to
develop new ways to continually affirm platform and sample integrity in
the absence of biology-based means. For example, what could be used for
a positive control material, given the scarcity and dissimilarity of
cancer samples? Method reproducibility is an issue, as mass spectrometry
has been developed for very high precision in species detection (M/Z)
but not in molecular concentration measurement (peak intensity). In
part, for example, assay variability could arise from potential
heterogeneity of molecules within a spot, which is why SELDI-TOF employs
multiple desorptions from different positions within a spot. Gordon
Whiteley and coworkers at NCI have undertaken a systematic study of
analytical variables — such as sample freeze-thaw effects, dilution
buffers, etc. — to understand sources of variation in SELDI-TOF. He
has reduced inter-assay peak height variability to about 11% or better
(coefficient of variation, major peaks) (12).
Mass spectrometry has taken a role
in the discovery of more traditional biomarkers as well. Zhang and
coworkers used SELDI-TOF to identify molecules potentially contributory
to ovarian cancer (13). They collected samples from cancer patients and
control subjects at five institutions in three countries. Biomarker
discovery was performed separately at two of the sites, and the results
were compared to obtain three consistent biomarkers that distinguished
cancer from non-cancer spectra. The discovery process was extensive:
serum samples were separated by anion exchange and pH elution into six
fractions, each of which was run in triplicate on four different chips
with different functional surfaces (Figure 1). Thus, each sample was run
on 72 spots, which provided an in-process confirmation of sorts.
The peaks most likely to
distinguish disease from control spectra were selected using a
commercial statistical algorithm (ProPeak; 3Z Informatics, Charleston,
South Carolina, USA). The three biomarkers discovered with this approach
had M/Z values of 3272 (identified as a fragment of inter-a
trypsin, up-regulated in cancer); 12,828 (a form of transthyretin,
down-regulated in cancer) and 28,043 (apolipoprotein A1, down-regulated
in cancer). These biomarkers then were validated using SELDI-TOF on
independent samples from four of the sites. In all, 503 samples were
used in discovery and mass spectrometry validation.
The validation studies showed that
the three-biomarker panel could classify an independent set of 35 Stage
I/II cancer samples and 63 healthy control samples with 97% specificity
and 74% sensitivity, compared to 97% specificity and 54% sensitivity
when CA-125 alone was used to classify. (The three biomarker set used in
conjunction with CA-125 showed no improvement in classification ability
over the biomarker set alone.)
These results suggest that the
three-biomarker panel provides early detection of ovarian cancer better
than CA-125, but not as well as required for a screening test. The
three-biomarker panel would have greater value if used in conjunction
with another complementary test.
A major advantage of isolating
discrete biomarkers is that immunoassays or other biomolecule-specific
assays can be developed for their detection. Immunoassays are performed
routinely in the clinical lab on automated platforms with high
throughput and, as such, are more economical and practical than
SELDI-TOF in its present form. However, development of sufficiently
sensitive immunoassays would be required. Some molecular isoforms cannot
be distinguished by immunoassays, in which case mass spectrometry or
another sensitive method would have to be used. The new ovarian
cancer-specific biomarkers have been described by Zhang et al., as acute
phase reactants; the specific relevance of these molecules to ovarian
cancer is not clear. More studies would be needed to validate the
discriminating nature of these markers and their utility when used with
other tests.
Lysophosphatidic acids (LPAs), a
category of phospholipids, were suspected to be related to ovarian
cancer because of their role in stimulating cell proliferation (14). In
1998, Xu and coworkers discovered that LPA could be an indicator for
late Stage I ovarian cancer (15), an observation they confirmed in a
subsequent study with the Moffitt Cancer Center (Tampa, Florida, USA).
In that study (16), Sutphen and coworkers analyzed blood plasma samples
from 45 confirmed ovarian cancer patients and 27 control subjects for 22
types of lysophospholipids (LPLs) including 10 types of LPA. The
experimental protocol, as summarized in Figure 2, is fairly elaborate,
and definitely is considered biomarker discovery rather than assay
development. For instance, after collection, whole blood was immediately
chilled and shipped to one location for centrifugation within 16–28 h
after draw — a process that is not routine in the reference
laboratory. Furthermore, samples were extracted and analyzed for lipid
fractions involving organic extraction and solvent evaporation steps
(Figure 2) that are not automated and would render the assay in its
current form unsuitable for high-throughput testing. The meticulously
extracted samples were analyzed with electrospray mass spectrometry —
a technique that is 10 times more sensitive than the gas chromatography
the researchers previously used (16). The results from this limited
study were impressive: the combined plasma concentrations of two LPA
categories, 16:0-LPA and 20:4 LPA, could classify cancer and non-cancer
samples with 93% accuracy at a 0.62 mmol/L cutoff (Figure 3). The two
lipid biomarkers show promise for early detection, as they could
classify early-stage (I/II), as well as later-stage disease, with high
accuracy (Figure 3). Even total LPA, at a 1.5 mmol/L cutoff, could
classify cancer and non-cancer specimens with 92% accuracy, suggesting
that lipid subfractionation might not be needed in the final assay
configuration. The authors observed that total LPA showed a significant
decrease in longitudinal samples collected from 22 patients before and
after operation, supporting that LPA is associated with the tumor.
Yet, several hurdles must be
overcome before laboratories can seriously consider using LPA as a
biomarker for ovarian cancer. First, the laboratory analysis must be
streamlined and automated; this is more readily achievable once the
definitive lipid biomarkers are identified since the procedure could be
targeted toward those markers alone. Laboratorians are not keen on
organic extraction; that step would have to be superseded. Next, the
reproducibility of the method has to be established and some questions
answered — will sample collection and transportation affect the
results, and will the LPA cutoff threshold bear out in larger studies?
Finally, the assay must be done on a statistically significant number of
case and control specimens that are independent (i.e., not used to
establish the threshold). This study, conducted on samples from multiple
institutions, would give a better indication of the assay's performance
when put to actual use.
Summary
The quest for an assay to detect ovarian cancer has advanced analytical
technology in several complementary areas, including patterning
software, mass spectrometry and lipid analysis. Approaches have ranged
from rote biomarker discovery to combinations of cutting-edge
technologies. Mass spectrometry pattern analysis has alienated some
traditional researchers who claim a biomarker should be identified
before it can be used (17). Consider aspirin, which was sold as a
painkiller for seven decades before its mode of action was understood
(18). Does that mean it should not have been used? This author feels
that pattern analysis has potential benefits that are compelling enough
that, once properly validated on a robust, reproducible platform, it
could have a role in improving healthcare. The challenge now is to
validate these assays and adapt them to routine, large-scale use.
A major hurdle to validation is the
scarcity of well-characterized specimens. Assay validation requires
significantly more samples than have been used in most studies to date;
ideally, 500 to 1000 cancer specimens and a comparable number of control
specimens from multiple institutions should be used. Archived ovarian
cancer specimens are scarce, especially from early-stage disease, so
understandably they only will be available to validate the most
promising methods. This poses a hardship on assay development, which
requires positive samples. New trials with prospective sample collection
will take a long time because ovarian cancer is a relatively rare
disease and patients must be followed up with to confirm their disease
state. Despite the hurdles, the independent, multi-marker techniques
under development could perform synergistically in cancer detection.
This gives hope that a definitive test will not be far off.
References
1. American Cancer Society: Cancer Facts & Figures 2005. Available
on-line at http://www.cancer.org/docroot/STT/stt_0.asp.
Last accessed Feb. 13, 2005.
2. R. Zurawski, H. Orjaster, A.
Andersen and E. Jellum, Int. J. Cancer 42, 677–680
(1988).
3. American Cancer Society: What
Are The Risk Factors for Ovarian Cancer? Available on-line at http://www.cancer.org/docroot/CRI/content/
CRI_2_4_2X_What_are_the_risk_factors_for_ovarian_cancer_33.asp.
Last accessed Feb. 13, 2005.
4. Mid-Atlantic Cancer Genetics
Network. Available on-line at http://www.macgn.org/factsheets/Hereditary%20Ovarian%20Cancer.pdf.
Accessed February 13, 2005.
5. E.F. Petricoin, A.M. Ardekani,
B.A. Hitt et al., Lancet 359, 572–577 (2002).
6. Mass Spectrometry Resource,
School of Chemistry, University of Bristol. Accessed February 14, 2005.
7. W. Zhu, X. Wang, Y. Ma et al., Proc.
Nat. Acad. Sci. 100, 14666–14671 (2003).
8. Ovarian cancer spectra obtained
by National Cancer Institute investigators. Available on-line at http://home.ccr.cancer.gov/ncifdaproteomics/ppatterns.asp.
Accessed February 14, 2005.
9. E. Check, Nature 429,
496–497 (2004).
10. J.M. Sorace and M. Zhan, BMC
Bioinformatics 4, 24–36 (2003).
11. Letter from FDA to Correlogic
Systems Inc., July 12, 2004. Available on-line at http://www.fda.gov/cdrh/oivd/letters/071204-correlogic.html.
Accessed February 14, 2005.
12. Gordon Whiteley, personal
communication, June, 2004.
13. Z. Zhang et al., Cancer
Research 64, 5882–5890 (2004).
14. Y. Xu, X.J. Fang, G. Casey and
G.B. Mills, Biochem J. 309, 933–940 (1995).
15. Y. Xu, Z. Shen and D.W. Wiper, JAMA
280, 719–723 (1998).
16. R. Sutphen et al., Cancer
Epidemil. Biomarkers Prev. 13, 1185–1191 (2004).
17. E.P. Diamandis, JNCI
95, 489 (2003).
18. See http://www.bayeraspirin.com/questions/asp_history.htm.
Accessed February 16, 2005.
Conan K.N. Li is chief
technology officer at VaxDesign Corporation in Orlando, Florida, USA. He
can be reached at cli@vaxdesign.com
; Tel. 407-249-3682.
|
