| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Editorials |
Department of Laboratory Medicine, Warren Magnuson Clinical Center, NIH, Bldg. 10, Room 2C-407, Bethesda, MD 20892-1508, Fax 301-402-1885, E-mail ghortin{at}mail.cc.nih.gov
This issue of Clinical Chemistry contains a report describing interlaboratory comparison of a prostate cancer test based on profiling of serum proteins by mass spectrometry (1). This report is relevant to the recent controversy regarding the diagnostic potential and reliability of this approach to the diagnosis of cancer or other diseases (2)(3)(4)(5)(6)(7)(8). This controversy has raised questions regarding whether mass spectrometric profiling of proteins can achieve standards of reproducibility and performance that are expected of clinical tests. Semmes et al. (1) examine whether different laboratories can achieve equivalent results on split specimens. These efforts are commendable starts to issues of standardization, calibration, and quality control (QC), all important elements in the transition of this technology from the discovery phase to clinical application. However, their report also exemplifies how initial efforts to apply this technology have not yet met the desired standards of clinical laboratory practice.
Although proteomic profiling has provided breakthroughs in the discovery of new disease markers, initial discovery methods generally have been poorly suited for clinical applications. Marker discovery methods typically have not incorporated principles applied to existing clinical laboratory applications of profiling methods, such as serum protein electrophoresis, amino acid analyses, or tandem mass spectrometric screening for inborn errors. Some lessons from clinical experience include the importance of (a) use of internal standards for mass spectrometry, (b) identifying measured components, (c) developing standards for calibration and QC, (d) identifying peaksets in spectra, and (e) applying established standards for method evaluation (9), e.g., measures of reproducibility, detection limits, linearity, and recovery; evaluation of calibration curves, potential interferences, reference intervals, and peak characteristics; and separations for profiling methods.
Proteomic profiling has been presented as a "new paradigm for diagnosis", and spectra are often described as "signatures", "proteomic features", or "fingerprints". This nomenclature reflects the fact that the components being measured and their quantitative relationships are not known. Although this paradigm may speed marker discovery, it prohibits the rigorous validation, standardization, and quality assurance demanded for clinical assays. Proteomic profiling will progress toward clinical practice as laboratories start analyzing quantifiable peaks representing known components rather than signatures and fingerprints. Laboratories actually do not analyze complete profiles for diagnostic applications, but only a panel of 
3–10 peaks selected from the total profile, and this represents an analysis of modest complexity. There is no fundamental reason that current laboratory standards cannot be applied to mass spectrometric profiling. Some recommendations on how to achieve this are listed in Table 1
.
|
There is a need to evaluate preanalytical issues more extensively. Studies should address patient preparation (such as fasting vs nonfasting or diurnal variation). Optimal materials and methods for sample collection and processing should be determined. Even selection of appropriate collection tubes may be important, as collection tubes can shed a variety of polymeric components detected in mass spectra (10). A reasonable goal is for collection procedures to minimize variability of results. Evaluation of specimen stability and criteria for specimen rejection are needed.
For the analytical process, both chromatographic steps and mass spectrometry need to be controlled. As noted by Semmes et al. (1), calibration of resolution, detection sensitivity, and mass accuracy are essential. Process control by automation of sample preparation is desirable. Clinical mass spectrometric assays generally use internal standards to correct for variable ionization and suppression (11)(12)(13). Internal standards might be either known added components or internal constituents assayed by independent methods such as immunoassay.
There are several reasons to determine sequence identities of peaks selected as diagnostic markers, and progress is being achieved in these efforts. Identification of peaks provides information about their potential origin and physiology. It identifies peaks representing different charge states, ion complexes, or molecular forms of the same component. Identifying components advances the development of reference materials for calibration, recovery experiments, and quality assurance.
Method evaluations should include standard measures expected of any clinical method (8), such as precision, detection limits, linearity, recovery, reference intervals, and potential interferences. Ideally, these evaluations require purified components for addition experiments. However, even if purified components are not available, studies such as detection limits and linearity evaluations still can be performed by mixing specimens with high and low concentrations.
Identifying materials and approaches for QC are important elements of clinical method development. Semmes et al. (1) addressed this issue, albeit with some shortcomings. QC programs should include at least two concentrations of QC material representing high and low specimens, and separate materials should be used for calibration and QC. Between-instrument and between-laboratory comparisons are important elements of reproducibility. Reference intervals might be determined for the peak ratios between specific components or vs an internal standard. Consideration should be given to potential interferences from sources such as hemolysis. Evaluation of lipemia might compare pre- vs postprandial specimens from the same individual.
Postanalytical components, the processing of data from mass spectrometric protein profiling, have generated considerable controversy (2)(3)(4)(5)(6)(7)(8). An interesting finding of Semmes et al. (1) was that diagnostic algorithms began to fail if analytical resolution or mass calibration deteriorated. This suggests a need to perform a sensitivity analysis for precision characteristics such as mass accuracy and peak ratio reproducibility to identify necessary performance characteristics. Care should be taken in diagnostic application of peaks with mass/charge <1200. In this region, there are high background signals from matrix. Candidate peaks for diagnostic applications ideally should have high signal intensities and reproducibility, and selected peaks should be divided equally among peaks that increase and decrease with disease to minimize effects of variation in absolute signal intensities. Semmes et al. (1) describe the better reproducibility achieved with measurement of high intensity vs weak peaks, as expected for any profiling method. Some of the challenges for any profiling method are, from the analytical perspective, to optimize peak intensities and resolution and, from the postanalytical perspective, to process data for optimal peak measurements.
Developing a valid diagnostic method depends on reliable classification of specimens that serve as training and evaluation sets for diagnostic algorithms. Adequate training of neural networks generally has been considered to require training sets with a sample size at least 5–10 times the number of data components analyzed (14). This places practical limits on the number of peaks or individual data points that can be analyzed for diagnosis. Training and evaluation specimens need to be representative of the population that will be tested; use of optimal or highly selected specimens is not likely to yield a realistic evaluation of diagnostic performance. Selection of optimal specimens, as in the study by Semmes et al. (1), prevents interpretation of how well their results reflect performance on unselected specimens, and this is identified as a future component of their studies.
Once a method has been established, the diagnostic algorithm must remain constant. Neural networks have the capacity for ongoing training, theoretically with continuous improvement. However, there are hazards of adding misclassified specimens to the database and of deteriorating performance with extended training on a fixed dataset (15). In addition, it would be difficult to QC an evolving algorithm. Any diagnostic algorithm should provide a quantitative value such as a probability score rather than a simple positive or negative value. Separate QC should be performed on the algorithm with use of fixed datasets to verify stable performance every day of use. Quantitative values allow assessment of best cutoff values by use of traditional tools such as ROC curves. Different cutoffs may be appropriate for different populations or for screening vs diagnostic applications.
Discoveries of potential new markers for cancer and other diseases by mass spectrometric protein profiling justifiably have created great excitement (2)(3)(4)(5). There also has been some disappointment and impatience that discovery methods have not been applied immediately to clinical diagnosis (8). This probably reflects a lack of appreciation of the many steps, such as evaluation of interlaboratory performance as in the study by Semmes et al. (1), from method discovery to clinical practice. As for any new clinical laboratory method, there is a need to develop and achieve standards of practice for mass spectrometric protein profiling that assure achievement of the quality of results that patients deserve.
References
The following articles in journals at HighWire Press have cited this article:
|
|
 |
|
  G. M. Fiedler, A. B. Leichtle, J. Kase, S. Baumann, U. Ceglarek, K. Felix, T. Conrad, H. Witzigmann, A. Weimann, C. Schutte, et al. Serum Peptidome Profiling Revealed Platelet Factor 4 as a Potential Discriminating Peptide Associated with Pancreatic Cancer Clin. Cancer Res., June 1, 2009; 15(11): 3812 - 3819. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 L. A. Liotta and E. F. Petricoin Putting the "Bio" back into Biomarkers: Orienting Proteomic Discovery toward Biology and away from the Measurement Platform Clin. Chem., January 1, 2008; 54(1): 3 - 5. [Full Text] [PDF]
|
|
|
|
|
|
R. E. Banks Preanalytical Influences in Clinical Proteomic Studies: Raising Awareness of Fundamental Issues in Sample Banking Clin. Chem., January 1, 2008; 54(1): 6 - 7. [Full Text] [PDF]
|
|
|
|
|
|
M. J. Bennett Untargeted Metabolomic Analysis Hits the Target Clin. Chem., December 1, 2007; 53(12): 2037 - 2039. [Full Text] [PDF]
|
|
|
|
|
|
N. Seam, D. A. Gonzales, S. J. Kern, G. L. Hortin, G. T. Hoehn, and A. F. Suffredini Quality Control of Serum Albumin Depletion for Proteomic Analysis Clin. Chem., November 1, 2007; 53(11): 1915 - 1920. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Hundt, U. Haug, and H. Brenner Blood Markers for Early Detection of Colorectal Cancer: A Systematic Review Cancer Epidemiol. Biomarkers Prev., October 1, 2007; 16(10): 1935 - 1953. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 B. Lumbreras, M. Porta, and I. Hernandez-Aguado Assessing the social meaning, value and implications of research in genomics J Epidemiol Community Health, September 1, 2007; 61(9): 755 - 756. [Full Text] [PDF]
|
|
|
|
|
|
J. Albrethsen Reproducibility in Protein Profiling by MALDI-TOF Mass Spectrometry Clin. Chem., May 1, 2007; 53(5): 852 - 858. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
G. L. Hortin A New Era in Protein Quantification in Clinical Laboratories: Application of Liquid Chromatography-Tandem Mass Spectrometry Clin. Chem., April 1, 2007; 53(4): 543 - 544. [Full Text] [PDF]
|
|
|
|
|
|
G. M. Fiedler, S. Baumann, A. Leichtle, A. Oltmann, J. Kase, J. Thiery, and U. Ceglarek Standardized Peptidome Profiling of Human Urine by Magnetic Bead Separation and Matrix-Assisted Laser Desorption/Ionization Time-of-Flight Mass Spectrometry Clin. Chem., March 1, 2007; 53(3): 421 - 428. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
G. L. Hortin The MALDI-TOF Mass Spectrometric View of the Plasma Proteome and Peptidome Clin. Chem., July 1, 2006; 52(7): 1223 - 1237. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
G. L. Hortin, S. A. Jortani, J. C. Ritchie Jr, R. Valdes Jr, and D. W. Chan Proteomics: A New Diagnostic Frontier Clin. Chem., July 1, 2006; 52(7): 1218 - 1222. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. E. Banks, A. J. Stanley, D. A. Cairns, J. H. Barrett, P. Clarke, D. Thompson, and P. J. Selby Influences of Blood Sample Processing on Low-Molecular-Weight Proteome Identified by Surface-Enhanced Laser Desorption/Ionization Mass Spectrometry Clin. Chem., September 1, 2005; 51(9): 1637 - 1649. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. R. Master Diagnostic Proteomics: Back to Basics? Clin. Chem., August 1, 2005; 51(8): 1333 - 1334. [Full Text] [PDF]
|
|
|
|
|
|
A. Karsan, B. J. Eigl, S. Flibotte, K. Gelmon, P. Switzer, P. Hassell, D. Harrison, J. Law, M. Hayes, M. Stillwell, et al. Analytical and Preanalytical Biases in Serum Proteomic Pattern Analysis for Breast Cancer Diagnosis Clin. Chem., August 1, 2005; 51(8): 1525 - 1528. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
